Battlecards are trusted in inverse proportion to their age, and they age in weeks. After a competitor's release or your own pricing change, every card that mentions either is wrong — but updating the full set is a project, so reps quietly stop trusting them and improvise on calls instead. The improvised answers are worse than the stale cards.
The maintenance burden is the disease: every fact appears on several cards, and hand-editing N copies per change guarantees drift.
A scheduled Fleet workflow regenerates the battlecard set from single-source corpora — product facts, pricing, competitive notes — with a review pass that flags unsupported or stale claims and a PMM approval before the set replaces the old one.
genflows:
- name: battlecards
schedule: "0 9 * * 1" # Monday mornings
steps:
- {name: cards, prompt: "Regenerate this battlecard from current product facts, pricing, and competitive notes. Objection responses must cite the corpus.", corpus: ["docs/product/**/*.md", "pricing.md", "intel/notes/**/*.md"], for_each: "sales/battlecards/*.md", kind: report, out: card.md}
- {name: claims, prompt: "Flag claims unsupported by the corpus and facts that changed since the last revision.", depends_on: [cards], kind: review, out: flags.md}
- {name: pmm-ok, depends_on: [cards, claims], kind: approval, out: decision.md}
The fan-out regenerates each card from shared sources, so a pricing change propagates to every affected card in one run — and fingerprinting skips cards whose inputs didn't change, so the Monday run is usually cheap.
Single-source corpora → fan-out card regeneration → claims review → PMM approval. Reps get cards whose freshness is bounded by the schedule, with every claim traceable.
PMM and sales enablement teams maintaining battlecards across more than a couple of competitors — anywhere card count × change rate has outrun hand maintenance.
Approved cards are published artifacts with stable links — most teams link them from the CRM or enablement tool they already use. The links serve approved revisions, so a rep can't accidentally read a draft.
That's its own loop — see the competitive-intel-briefs workflow. The two compose: notes feed the brief and the cards from the same corpus.
One binary. Five minutes. See every agent, coordinate every handoff, and keep a full audit trail of what your fleet did.